4 ms·
> What we need is not a "I know what you did and here is more of that" suggestion engine I really agree on this. The Youtube homepage exemplifies it for me. It
by wvrvwwwe 10y ago
> What we need is not a "I know what you did and here is more of that" suggestion engine
I really agree on this. The Youtube homepage exemplifies it for me. It recommends only content the same as or very similar to what I've watched before, and only the most popular of that. I've watched a ton of conference talks, programming tutorials, etc. yet those never show up. It's mostly junk with clickbait titles from big channels.
I do like Youtube's recommendation algorithm for the side bar suggestions on a video page, though I get more sponsored-looking content now than in the past. It's been useful at least for finding music. And YT does have a page for improving homepage recommendations, which granted I haven't tried yet.
- pjc50 10y agoYoutube's home page is especially terrible. It seems to have very broad categories, and there's no really good way of pruning terrible (but popular) stuff out of it. You can say "not interested", then get some almost identical channel instead. Whereas the "related" sidebar while actually watching a video is great.
- funkyy 10y ago>And YT does have a page for improving homepage recommendations, which granted I haven't tried yet. Never hear of it. Can you share a link to it or a name of it?
- wvrvwwwe 10y agoThis is what I was thinking of: https://support.google.com/youtube/answer/6342839?hl=en https://support.google.com/youtube/answer/6342839?hl=en I didn't look closely at it the first time, but now that I have, it's not that useful.
- anexprogrammer 10y ago"I do like Youtube's recommendation algorithm for the side bar suggestions" You do? It does ok if I'm watching something technical or some reviews. It's laughably bad at music and wider interest content such as a documentary. If you hit a music track, 50% of the recommendations are the same damn track in degrading levels of quality, 30% the same band. The 10% that's left for other artists are normally a really bad connection to the mood. The last 10% lately is taken by awfully bad match, clearly sponsored, junk though never identified as sponsored. Most of the time I'd like suggestions in the same genre or mood of music. If I've just watched an upbeat rock track why not suggest other upbeat rock tracks? Last.fm radio was superb at this, for years, until CBS destroyed it. So it can't be that hard. I think the only times I use the sidebar is if the vid I landed on is terrible quality or a bad hit.
- wvrvwwwe 10y agoI don't know if this has anything to do with it, but I only seek out full albums. I've found that YT does pretty well for classic rock, metal, and electronica, and ok for older Jazz and classical music. I tried it out just now for an album I like. I get about 25% unrelated junk or stuff I've already watched, but the rest are also full albums that seem like fairly solid recommendations based on the album at hand (Jan Jelinek's "Loop Finding Jazz Records"). https://www.youtube.com/watch?v=6hIgBEXuQD8 https://www.youtube.com/watch?v=6hIgBEXuQD8
- anexprogrammer 10y agoInteresting. I very rarely search whole albums on YT but often rock and metal. Start with a random metal track in mind an I find I get a band mix suggestion, a nordic folk metal mix, 11 more tracks from the same band, 3 tracks I listened to very recently, 4 connected other artists and "5 CREEPIEST Old Found Footages - Recommended for you" as a perfect example of the crap being injected recently. They must notice I've never once clicked one. I do clear cookies and history periodically, so perhaps not having a multiple year history counts against.
- wvrvwwwe 10y ago> I do clear cookies and history periodically, so perhaps not having a multiple year history counts against. In that other link I posted YT suggests pruning your watched history to get better recommendations -- but that's a huge undertaking given the thousands of garbage videos that I've clicked out of in five seconds. Seems like it would be better if the algorithm only operated on data from videos a user manually "likes".